We propose a novel reinforcement learning approach to learning lattice codes for MIMO channels. We use the block error rate as a loss function to be minimized and compare the learnt lattices with those obtained from algebraic design methods for different SNR ranges. Our results indicate that our learnt lattices achieve close to optimal performance in some cases.

The design of lattice coset codes for wiretap channels is considered. Bounds on the eavesdropper's correct decoding probability and information leakage are first revisited. From these bounds, it is explicit that both the information leakage and error probability are controlled by the average flatness factor of the eavesdropper's lattice, which we further interpret geometrically. It is concluded that the minimization of the (average) flatness factor of the eavesdropper's lattice leads to the study of well-rounded lattices, which are shown to be among the optimal in order to achieve these minima. Constructions of some well-rounded lattices are also provided.